🪡 Orra (✨Alpha✨)

Move beyond simple Crews and Agents. Use Orra to build production-ready multi-agent applications that handle complex real-world interactions.

Orra coordinates tasks across your existing stack, agents and any tools run as services using intelligent reasoning — across any language, agent framework or deployment platform.

• 🧠 Smart pre-evaluated execution plans
• 🎯 Domain grounded
• 🗿 Durable execution
• 🚀 Go fast with tools as services
• ↩️ Revert state to handle failures
• ⛑️ Automatic service health monitoring
• 🔮 Real-time status tracking
• 🪝 Webhook result delivery

Read the launch blog post.

Coming Soon

• Agent replay and multi-LLM consensus planning
• Continuous adjustment of Agent workflows during runtime
• Additional language SDKs - Ruby, DotNet and Go very soon!

Table of Contents

• Installation
• How The Plan Engine Works
• Self Hosting
• Join Our Alpha Testing Community
• License

Installation

Prerequisites

• Docker and Docker Compose - For running the control plane server (powers the Plan Engine)
• Set up Reasoning and Embedding Models to power task planning and execution plan caching/validation

Setup Reasoning Models

Select between Groq's deepseek-r1-distill-llama-70b model or OpenAI's o1-mini / o3-mini models.

Update the .env file with one of these:

Groq

# GROQ Reasoning
REASONING_PROVIDER=groq
REASONING_MODEL=deepseek-r1-distill-llama-70b
REASONING_API_KEY=xxxx

O1-mini

# OpenAI Reasoning
REASONING_PROVIDER=openai
REASONING_MODEL=o1-mini
REASONING_API_KEY=xxxx

O3-mini

# OpenAI Reasoning
REASONING_PROVIDER=openai
REASONING_MODEL=o3-mini
REASONING_API_KEY=xxxx

Setup Embedding Models

Update the .env file with:

# Execution Plan Cache and validation OPENAI API KEY
PLAN_CACHE_OPENAI_API_KEY=xxxx

1. Install Orra CLI

Download the latest CLI binary for your platform from our releases page:

# macOS
curl -L https://github.com/ezodude/orra/releases/download/v0.2.1/orra-macos -o /usr/local/bin/orra
chmod +x /usr/local/bin/orra

# Linux
curl -L https://github.com/ezodude/orra/releases/download/v0.2.1/orra-linux -o /usr/local/bin/orra
chmod +x /usr/local/bin/orra

# Verify installation
orra version

→ Full CLI documentation

2. Get Orra Running

Clone the repository and start the control plane:

git clone https://github.com/ezodude/orra.git
cd orra/controlplane

# Start the control plane
docker compose up --build

How The Plan Engine Works

The Plan Engine powers your multi-agent applications through intelligent planning and reliable execution:

Progressive Planning Levels

1. Base Planning

Your agents stay clean and simple (wrapped in the Orra SDK):

Python

from orra import OrraAgent, Task
from pydantic import BaseModel

class ResearchInput(BaseModel):
    topic: str
    depth: str

class ResearchOutput(BaseModel):
    summary: str

agent = OrraAgent(
    name="research-agent",
    description="Researches topics using web search and knowledge base",
    url="https://api.orra.dev",
    api_key="sk-orra-..."
)

@agent.handler()
async def research(task: Task[ResearchInput]) -> ResearchOutput:
    results = await run_research(task.input.topic, task.input.depth)
    return ResearchOutput(summary=results.summary)

JavaScript

import { initAgent } from '@orra.dev/sdk';

const agent = initAgent({
  name: 'research-agent',
  orraUrl: process.env.ORRA_URL,  
  orraKey: process.env.ORRA_API_KEY
});

await agent.register({
  description: 'Researches topics using web search and knowledge base',
  schema: {
    input: {
      type: 'object',
      properties: {
        topic: { type: 'string' },
        depth: { type: 'string' }
      }
    },
    output: {
      type: 'object',
      properties: {
        summary: { type: 'string' }
      }
    }
  }
});

agent.start(async (task) => {
  const results = await runResearch(task.input.topic, task.input.depth);
  return { summary: results.summary };
});

Features:

• AI analyzes intent and creates execution plans that target your components
• Automatic service discovery and coordination
• Parallel execution where possible

2. Production Planning with Domain Grounding

# Define domain constraints
name: research-workflow
domain: content-generation
use-cases:
  - action: "Research topic {topic}"
    capabilities: 
      - "Web search access"
      - "Knowledge synthesis"
constraints:
  - "Verify sources before synthesis"
  - "Maximum research time: 10 minutes"

Features:

• Full semantic validation of execution plans
• Capability matching and verification
• Safety constraints enforcement
• State transition validation

3. Reliable Execution

# Execute an action with the Plan Engine
orra verify run "Research and summarize AI trends" \
  --data topic:"AI in 2024" \
  --data depth:"comprehensive"

The Plan Engine ensures:

• Automatic service health monitoring
• Stateful execution tracking
• Built-in retries and recovery
• Real-time status updates
• Webhook result delivery

Explore Examples

• 🛒 E-commerce AI Assistant (JavaScript) - E-commerce customer service with a delivery specialized agent
• 👻 Ghostwriters (Python) - Content generation example showcasing how to use Orra with CrewAI 🆕🎉
• 📣 Echo Service (JavaScript) - Simple example showing core concepts using JS
• 📣 Echo Service (Python) - Simple example showing core concepts using Python

Docs and Guides

• Rapid Multi-Agent App Development with Orra
• What is an Agent in Orra?
• Orchestrating Actions with Orra
• Domain Grounding Execution
• Core Topics & Internals

Self Hosting

1. Storage: We use BadgerDB to persist all state
2. Deployment: Single-instance only, designed for development and self-hosted deployments

Join Our Alpha Testing Community

We're looking for developers who:

• Are building multi-agent applications
• Want to help shape Orra's development
• Are comfortable working with Alpha software
• Can provide feedback on real-world use cases

Connect With Us:

• GitHub Discussions - Share your experience and ideas
• Office Hours - Weekly calls with the team

License

Orra is MPL-2.0 licensed.